AI companies inherited usage-based pricing from the infrastructure layer, then bolted it onto products that deliver value per outcome. The result is buyers who fear unpredictable bills, margins that evaporate when inference costs spike, and a model nobody on the buying committee can forecast.
The Problem
Token and usage pricing terrifies the buyer who has to forecast the bill
When the invoice scales with tokens consumed, the economic buyer cannot predict next quarter's spend, and unpredictable spend is the fastest way to lose a procurement approval. Finance blocks deals they cannot model, and champions stop expanding usage because they are afraid of a surprise bill. The pricing meant to feel fair and consumption-aligned instead caps adoption, because the person signing the contract optimizes for predictability over theoretical fairness.
Margins collapse when price is tied to a cost you don't control
If you price per token or per inference and your underlying model or GPU cost moves, your margin moves with it – and not in your favor. AI companies that pass through a thin markup on inference have no buffer when a provider raises rates or a workload shifts to a more expensive model. They discover at the worst possible moment that their gross margin is hostage to a vendor's pricing decision, and they cannot raise prices fast enough to recover.
Price is anchored to cost instead of the value delivered
Tokens, compute, and seats are inputs. The buyer cares about the outcome – resolved tickets, generated code shipped, documents processed, fraud caught. Pricing on inputs leaves enormous value uncaptured when a single inference replaces hours of human work, and it makes the product look expensive when usage is high but value per call is low. Cost-plus pricing in AI systematically misprices the product relative to what it is actually worth to the buyer.
Free-tier and PLG economics quietly bleed money at scale
A generous free tier that made sense when inference was cheap becomes a real cash drain when heavy free users run expensive models all day. AI companies running product-led growth often have no model for the unit economics of a free account, so a viral moment that should be a win turns into a margin emergency. Without packaging that gates the expensive capabilities behind paid plans, the self-serve funnel subsidizes users who will never convert.
How We Help
We start by separating what your product costs to run from what it is worth to the buyer – two numbers most AI companies have collapsed into one. In the first 30 days we model your true unit economics per workload, per model, and per customer segment, then research what outcome the buyer is actually paying for. This tells us where cost-based pricing is leaving money on the table and where it is exposing you to margin risk you cannot control.
Strategy development designs the model and packaging. We identify a value metric that tracks the outcome the buyer cares about – not raw tokens – and build pricing around it, often a hybrid of a predictable platform fee plus metered usage so the economic buyer can forecast while heavy users still pay for what they consume. We design tiers and packaging that gate expensive capabilities appropriately, protect margin against inference-cost swings with floors and buffers, and give procurement a structure they can approve. This is core product strategy work, so we connect pricing tightly to packaging and roadmap.
Execution operationalizes the new model. We build the pricing logic, the migration plan for existing customers – which is where most repricing efforts break – and the sales and self-serve assets that explain the new model without scaring the buyer. We test price points and packaging with real prospects before a full rollout, and we sequence the change so existing accounts move without churn spikes. We work with your finance and product teams so the model holds up in a CFO's spreadsheet and in the product UI.
Measurement tracks margin per segment, forecastability for the buyer, expansion revenue, and win-rate change at the new price. Pricing strategy for AI companies works when the buyer can predict their bill, your gross margin survives an inference-cost spike, and price climbs with the value you deliver instead of the compute you burn.
What we deliver
- Unit-economics model per workload, model, and customer segment
- Value-metric definition that tracks buyer outcomes, not raw token usage
- Hybrid platform-plus-usage model with margin floors against inference-cost swings
- Packaging and tiering that gates expensive capabilities and fits procurement
- Migration plan and sales enablement to reprice existing customers without churn
AI companies keep pricing on tokens because that is what their costs look like. But the buyer pays for outcomes, not inputs – and the economic buyer pays even more for predictability. Win on a value metric the buyer can forecast, and protect margin so an inference-cost spike does not become a crisis.
Our Methodology
Our pricing strategy build for AI and ML companies runs as a 90-day program install. Phase one is the dual model – true unit economics per workload and model on one side, buyer value and willingness to pay on the other. We quantify where current pricing leaves value uncaptured and where it exposes margin to costs you do not control.
Phase two designs the model and packaging. We define a value metric tied to outcomes, build a hybrid of predictable platform fee plus metered usage, set margin floors and buffers against inference-cost volatility, and design tiers that gate expensive capabilities. We pressure-test price points and packaging with real prospects before committing.
Phase three operationalizes the change: pricing logic, a migration plan for existing customers, and sales and self-serve enablement that makes the new model legible to procurement. Unlike consultants who hand over a pricing deck and leave, we build the model, validate it with the market, and run the migration that determines whether a reprice succeeds or triggers churn.
How We Work
Initial engagements run 3 to 5 months because pricing work moves from modeling to market validation to a carefully sequenced rollout, and the migration of existing customers cannot be rushed. The first 30 days build the unit-economics and willingness-to-pay model. Days 31 to 60 design the model, packaging, and margin protection and test with prospects. Days 61 to 120 operationalize pricing logic, the migration plan, and sales enablement.
Our team includes a pricing strategist who owns the model, an analyst who builds the unit economics and willingness-to-pay research, and an operator who handles packaging, migration, and enablement. From your side we need finance for cost and margin data, product for packaging and roadmap alignment, and sales for buyer and deal-objection input. We handle modeling, research, design, and rollout planning.
Weekly working sessions move the model from analysis to design to validated rollout. A pre-launch review pressure-tests pricing against real deals and the migration plan against your existing base. Most AI companies have a validated model within 60 days and a sequenced rollout underway by 90, with margin and forecastability improvements measurable as customers move onto the new model over the following quarter.
- Gross-margin protection against inference and model-cost swings
- Forecastable pricing that clears procurement faster
- Expansion revenue from a value metric that grows with usage value
- Win-rate and deal-size improvement at the repriced model
If your ai / machine learning company needs pricing strategy leadership, we should talk.
Expand your marketing team output with our experts
Let us take a custom approach to your growth goals by assembling and leading the best-in-class marketing team to support your next stage.
Frequently asked questions
How much does a pricing strategy engagement cost for an AI company?
Most AI pricing engagements run between $30K and $75K for a focused 3-to-5-month project, depending on the complexity of your unit economics and how large a migration the reprice requires. That is far less than the margin you leave on the table with cost-plus token pricing or the revenue you lose to a model procurement cannot approve.
How long before a repricing produces results for an AI company?
A validated pricing model is usually ready within 60 days, and a sequenced rollout is underway by 90. Margin and forecastability gains show up as customers move onto the new model, which happens over the following quarter for new deals and longer for migrated accounts. The full revenue and margin impact lands once most of your base is on the new model, typically within two to three quarters of launch.
How does the pricing team integrate with our finance and product staff?
We work most closely with finance for cost and margin data and with product for packaging and roadmap alignment, since pricing and packaging are inseparable. Sales gives us buyer objections and deal context. We run the modeling, research, and design and bring your teams in at decision points so the model holds up in a CFO's spreadsheet and ships cleanly in the product.
What makes Winston Francois different from a traditional pricing consultancy?
Most pricing consultancies deliver a deck of recommendations and leave the hard part – migration and rollout – to you. We build the model, validate it with real prospects, and run the migration plan that determines whether a reprice grows revenue or triggers churn. We also bring operator context on AI unit economics, so the model accounts for inference-cost volatility rather than assuming stable costs.
How do you measure ROI from a pricing strategy engagement?
We measure gross margin per segment, win rate and deal size at the new price, expansion revenue from the value metric, and how predictably the buyer can forecast their bill. The headline is margin and net revenue improvement against the old model, net of any churn during migration. Most AI companies can see the margin and forecastability change within a quarter of moving customers onto the new model.
Should we move off usage-based pricing entirely?
Usually not entirely. Pure usage pricing scares the economic buyer, but pure flat pricing leaves money on the table from heavy users. For most AI companies the answer is a hybrid – a predictable platform fee the buyer can forecast plus metered usage tied to a value metric – which protects forecastability while still capturing value from high-usage accounts. The right mix depends on your unit economics and buyer, which we model in the first 30 days.
What type of AI company is the right fit for a pricing engagement?
Companies whose pricing is anchored to tokens, compute, or seats while their value is an outcome, and any AI company watching inference costs erode margin or losing deals to unforecastable bills. Post-product-market-fit AI companies preparing to scale revenue get the most from it. The first step is a unit-economics review to find where your current model is leaking margin or capping adoption.
